When synthetic images of real people must be labelled
If you publish image, audio or video content that resembles real people, places or events and has been artificially generated or manipulated, Article 50(4) requires you to disclose that it is artificial. This is a visible disclosure to people — separate from the machine-readable marking in 50(2).
What the law requires
Article 50(4) applies to deployers, not just providers: whoever publishes the deepfake carries the obligation to disclose it.
The definition is broader than the popular sense of the word. It is not limited to face swaps of celebrities — it covers content resembling real persons, objects, places, entities or events that would falsely appear authentic.
There is a narrower disclosure route where the content is part of an evidently artistic, creative, satirical or fictional work: the disclosure must still exist but may be made in a way that does not spoil the work.
AI-generated stock-style imagery of people who do not exist is generally not a deepfake, because it does not resemble a specific real person. It is still AI-generated content for the purposes of 50(2) marking.
What our scanner looks for
- Media on your pages that appears to depict real people or recognisable places alongside signals of synthetic generation.
- Whether any disclosure text accompanies that media — in a caption, adjacent copy, or the page's structured content.
- This check is reported as a warning or as needing review, never as an automated failure. Deciding whether something is a deepfake requires knowing what is real, which a crawler cannot establish.
How to fix it
- 1Label synthetic media where it appears, not in a footer: a caption reading "AI-generated image" directly under the asset.
- 2For creative or satirical work, keep the disclosure present but proportionate — end credits, an artist's note, or a collapsed caption.
- 3Keep a record of which published assets are synthetic. The hard part of responding to a complaint is usually not the labelling but knowing which files need it.
Deadlines and penalties
Enforceable since 2 August 2026, under the same Art. 99(4) penalty ceiling of €15 million or 3% of worldwide annual turnover. Note that from 2 December 2026 the Digital Omnibus also brings non-consensual intimate imagery and CSAM generation into the Art. 5 prohibitions — those are banned outright, not merely subject to disclosure.
Free, no account needed. We'll show you every page where this applies, with the evidence.
Common questions
Does an AI-generated image of a person who doesn't exist need a deepfake label?
Generally no, because it does not resemble a specific real person. It is still AI-generated content and so still falls under the Art. 50(2) machine-readable marking obligation.
Is a small watermark in the corner enough?
Article 50(4) requires disclosure, and a clear visible mark can satisfy it. The test is whether a person would actually notice and understand it — a faint mark that blends into the image does not.
Related
This page describes technical indicators and the obligations they relate to. It is not legal advice. A website scan detects indicators, not legal compliance — which is why our reports say “no issues detected” and never “compliant”.